Recently, I started to get in touch with reinforcement learning, bought some reference books, and read a lot of articles written by great gods on the Internet. Now I summarize the better information I have seen on the Internet, which is convenient for my own reference in the future, and I hope it can help some beginners. Because my English is not very good, the materials I read are basically in Chinese. If you have a certain foundation, it is recommended to read more blogs and papers of foreign great gods.
The first to know the column-- Reinforcement Learning Knowledge Lecture Hall
The content of this column is relatively comprehensive. The two major branches in the field of reinforcement learning, the value function method and the policy gradient method, are introduced in detail. At the same time, some python-based programming examples are given. After reading this column carefully, You can basically master most of the knowledge of reinforcement learning. The columnist also published the book "Intensive Learning in Simple Strategies". The content of the book is basically the articles in the column. Students who like to read books and study can buy a copy, which can be regarded as supporting the author. After all, originality is not easy.
Secondly, I recommend a knowing column-- Mofan
This column focuses on programming practice, the principle is not in-depth, there is no complicated mathematical derivation, it is very suitable for beginners to quickly understand different reinforcement learning methods, but I personally feel that some things are not very clear, and you need to consult other materials for help understand. It is recommended to visit the author's personal website-- Mofan python , the website covers the content of the column, and the author also puts the source code in github, which is simply a conscience author. At the same time, there are other machine learning content on the website. I really want to give the author 100,000 likes!
Recommend another know-how column-- smart unit
This column has a lot of introductions to the more cutting-edge content in the field of reinforcement learning research. At the same time, I think the articles about DQN in it are very well written and easy to understand!
After reading the above three columns carefully, it is basically the same. If you want to follow the frontier fields, you should read more foreign papers.
The following are some of the problems encountered during personal study and the materials consulted:
About on-policy and off-policy
https://blog.csdn.net/mmc2015/article/details/58021482
https://blog.csdn.net/u013615687/article/details/71055870
About Action Strategy
https://blog.csdn.net/hanlin_tan/article/details/62078935
https://blog.csdn.net/wangweiran1/article/details/49855959
https://www.cnblogs.com/blueyyc/p/5544752.html
About Importance Sampling
https://blog.csdn.net/baimafujinji/article/details/53869358
https://blog.csdn.net/u011332699/article/details/74298555
About DQN
https://blog.csdn.net/itplus/article/details/9361915
https://blog.csdn.net/songrotek/article/details/50951537
https://blog.csdn.net/qq_32231743/article/details/72809101
https://blog.csdn.net/Charel_CHEN/article/details/77408050?ref=myread
https://zhuanlan.zhihu.com/p/21421729 This article is highly recommended, it helps me understand how the Q-value neural network is updated.